Exposure Patterns Driving Ebola Transmission in West Africa: A Retrospective Observational Study

Agua-Agum, Junerlyn; Ariyarajah, Archchun; Aylward, Bruce; Bawo, Luke; Bilivogui, Pepe; Blake, Isobel M.; Brennan, Richard J.; Cawthorne, Amy; Cleary, Eilish; Clement, Peter; Conteh, Roland; Cori, Anne; Dafae, Foday; Dahl, Benjamin; Dangou, Jean-Marie; Diallo, Boubacar; Donnelly, Christl A.; Dorigatti, Ilaria; Dye, Christopher; Eckmanns, Tim; Fallah, Mosoka; Ferguson, Neil M.; Fiebig, Lena; Fraser, Christophe; Garske, Tini; Gonzalez, Lice; Hamblion, Esther; Hamid, Nuha; Hersey, Sara; Hinsley, Wes; Jambei, Amara; Jombart, Thibaut; Kargbo, David; Keita, Sakoba; Kinzer, Michael; George, Fred Kuti; Godefroy, Beatrice; Gutierrez, Giovanna; Kannangarage, Niluka; Mills, Harriet L.; Moller, Thomas; Meijers, Sascha; Mohamed, Yasmine; Morgan, Oliver; Nedjati-Gilani, Gemma; Newton, Emily; Nouvellet, Pierre; Nyenswah, Tolbert; Perea, William; Perkins, Devin; Riley, Steven; Rodier, Guenael; Rondy, Marc; Sagrado, Maria; Savulescu, Camelia; Schafer, Ilana J.; Schumacher, Dirk; Seyler, Thomas; Shah, Anita; Van Kerkhove, Maria D.; Wesseh, C. Samford; Yoti, Zabulon

Understanding when onward exposure occurred in the clinical course of infection in a primary case could inform the focus of interventions, for instance for early detection of cases or for improving the safety of funerals. We analysed the timing of non-funeral exposure events relative to the following clinical events of the named contacts: onset of symptoms, hospitalisation, and death. Here, we describe the analysis of the time from symptom onset to onward exposure; the analyses for hospitalisation and death were similar. We fitted (by maximum likelihood) a distribution to the observed delays between symptom onset and exposure. In brief (see section 1.8 in S1 Text for more detail), this likelihood accounted for (1) multiple exposures (when multiple exposures were reported, only one was likely to have led to infection, so exposures were weighted accordingly), (2) inaccurate recall of contacts, (3) inaccurate recall of dates of exposures, (4) errors in data entry. In order to address issues 2 to 4, we fitted a mixture of two offset lognormal distributions: the more peaked we regarded as the signal, and the broader we regarded as the noise (see section 1.8 in S1 Text). We then interpreted the signal distribution as our best estimate of actual infection times. Missing delays were imputed by random draws from the observed data. The analyses described above showed that the non-funeral exposure events were concentrated shortly after the onset of symptoms and around the day of death (see Results). We performed an additional analysis to further explore whether one of these two clinical events was